> On March 18, 2016, 6:36 p.m., Jiang Yan Xu wrote: > > configure.ac, line 923 > > <https://reviews.apache.org/r/44945/diff/3/?file=1304828#file1304828line923> > > > > Sorry this should have been a continued discussion on > > https://reviews.apache.org/r/44342/#comment184051 but anyways: > > > > Your response to my initial comments: > > > There's no AC_ARG_WITH. If the dependencies area available we will > > build the isolator and the operator can configure it at runtime. I don't > > think there's any need to disable this at build time since it is not > > enabled by default anyway. > > > > I am still favoring building XFS isolator based on > > `--with-xfs-isolator` (AC_ARG_ENABLE is probably better than AC_ARG_WITH) > > for the following reasons: > > > > > > 1. Explicitness: In general, if we build things solely based on the > > availability of headers, users can inadvertently build more things into the > > deployed binary which could result in larger binaries and perhaps other > > side effects (through bugs, etc.). > > 2. Preventing user mistake: I am thinking about how we would advise > > users on enabling this feature: > > 1. "The XFS isolator feature is disabled by default, to enable it in > > build, install these packages whose name could be different for different > > distros and if you haven't made mistakes in finding them, the feature will > > be built." vs. > > 2. "The XFS isolator feature is disabled by default, to enable it in > > build, install these packages (which could be named differently depending > > on your distro), if the dependencies are not satisfied, the build aborts > > with an message stating such error." The latter feels better to me. > > 3. Consistency: Other optional features of Mesos follow this pattern. > > Jiang Yan Xu wrote: > Sorry I meant `--enable-xfs-isolator`. > > Jie Yu wrote: > For consistency: I think the network isolator is using > `--with-network-isolator` (`AC_ARG_WITH([network-isolator],`), so it'e better > to be consistent and use 'AC_ARG_WITH' here. > > I agree with Yan that we should be more explicit, instead of relying on > headers.
My view is that unless there is a tradeof (performance, security, etc), then every optional feature should be available at runtime. The alternative imposes an undue burden on operators because they will have to repackage if they want to enable a feature. However, I'm not going to argue this any furthere here. I'll just guard this with ``AC_ARG_WITH`` (@xujyan is right this should have been ``AC_ARG_ENABLE``). - James ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/44945/#review124239 ----------------------------------------------------------- On March 18, 2016, 3:45 a.m., James Peach wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/44945/ > ----------------------------------------------------------- > > (Updated March 18, 2016, 3:45 a.m.) > > > Review request for mesos, Jie Yu and Jiang Yan Xu. > > > Bugs: MESOS-4828 > https://issues.apache.org/jira/browse/MESOS-4828 > > > Repository: mesos > > > Description > ------- > > Add autoconf tests for XFS project quotas. > > > Diffs > ----- > > configure.ac 8e4f03593df4a8ba13f00292963e351acc3f71c1 > > Diff: https://reviews.apache.org/r/44945/diff/ > > > Testing > ------- > > Make check. Manual verification. > > > Thanks, > > James Peach > >
